Machine learning is a subdiscipline of computer science focused on the capacities for computer systems to learn without being purposefully programmed to perform specific tasks. These capacities derive from pattern recognition and artificial intelligence (AI). The Capital One Machine Learning is a fast-paced research stream that will focus on using machine learning to develop algorithms involved in predictive data analytics using both supervised and unsupervised approaches.

In 2018, the stream will explore the state-of-art techniques in machine learning, data analytics and cybersecurity using recently developed perspectives, technologies and applications. This new FIRE stream will allow students who have interest in these fields to start tackling the critical challenges in our world with hands-on training, peer mentorship, and collaborative research.

Potential Research Questions in 2018

Machine Learning for Computer Vision and Visual Recognition

It is estimated that by 2019, 84% of the world's Internet traffic will be visual. Visual object recognition is enabling innovative systems like self-driving cars, semantic-based image retrieval, and autonomous unmanned systems. Deep learning added a huge boost to the already rapidly developing field of computer vision. However, the deep learning models these systems rely on can be difficult to design, train, and evaluate. With these pressing challenges, can we design an algorithm that efficiently learns to model deep neural networks?

Machine Learning for Natural Language Processing and Machine Translation

Every day, we are awash with text, from blogs, tweets, news, books, papers, and increasingly text from spoken utterances. Working with natural language is challenging given the inherent ambiguity and flexibility of human language. The recent development of encoder-decoder recurrent neural network architectures achieved state-of-the-art results compared to classical rule-based systems. Although effective, the neural machine translation systems still suffer some issues, such as slower training and inference speed, ineffectiveness in dealing with rare words, and sometimes failure to translate all words in the source sentence. With these issues, can we design a model that more completely represent linguistic ideas of a given text?

Machine Learning for Cybercrime and Fraud Detection

Detection of fraudulent activity in commercial transactions presents a significant opportunity potentially worth billions of dollars per year. Creating automated systems that can detect fraud is a natural machine learning problem. However, shifts in user behavior over time can create biases in the predictions that they make. A cybercriminal could potentially generate adversarial sequences to avoid flagging future transactions as fraudulent. Given historical transaction data, can we design an algorithm that learns to robustly detect fraudulent transactions?

Stream Sponsors

Capital One offers a broad array of financial products and services to consumers, small businesses and commercial clients in the U.S., Canada and the UK. Capital One is a major partner in the design and creation of the FIRE steam Capital One Machine Learning .

Datacamp offers interactive R and Python courses on topics in data science, statistics, and machine learning. Learn from a team of expert teachers in the comfort of your browser with video lessons and fun coding challenges. Datacamp is a sponsor in providing interactive video lessons for the FIRE steam Capital One Machine Learning.

Faculty Advisor
Dr. Dave Levin

Research Educator
Dr. Raymond Tu